Nonlinear inverse optimization for parameter estimation of commodity-vehicle-decoupled freight assignment

A systematic approach to estimate parameters from noisy priors is proposed for traffic assignment problems. It extends inverse optimization theory to nonlinear problems, and defines a new class of parameter estimation problems in the transportation literature for networks under congestion. The approach is used to systematically calibrate a new link-based variation of the STAN model which decouples commodity flows and vehicle flows. The models are tested on a small network and then a case study with real data from California statewide implementation. Cross-validation shows 15% CV of the RMSE.